Skip to content
DAO Labs

Why AI Is Only as Sharp as the Hands Holding It?

Malte Christensen

Malte Christensen

November 14, 2026

Legacy
How Claude processed 341 pages of lab data in 8 minutes, and what this human-in-the-loop way of working means for Social Miners.

MC.- When we broke down how Cody Tranbarger, aka Professor Oak on X, built Mithril Medicine into a one-person biotech acquired by Anthropic, one question immediately dominated the conversation: What was Claude actually doing behind the curtain? Please read the full explainer on Alexis’s article about it in our last blog entry.

It is easy to treat artificial intelligence like a magic black box. The reality inside Mithril Medicine’s operational logs are more empowering for anyone building in decentralized ecosystems.


Here are the three key moments from one week of preclinical work. Mithril’s technical report describes each request but never publishes the exact wording, so the prompts below are our own prompt recreation –for illustration purposes only– to show what each request likely looked like, not quotes from Mithril. Throughout the week, Claude could read the team’s email and files and save draft emails, but it sent nothing itself: a person reviewed and sent every message. 





 * 08:00 Minutes: Read and Sort
“Attached is the production CRO's 341-page data package for our 113-construct panel. Please read every page, share your observations and recommend a next step for each construct”.


A partner laboratory delivers a dense, 341-page technical dossier covering 113 antibody candidates under investigation. In eight minutes flat, Claude ingests the entire document, cross-references the biological targets, and maps out a concrete course of action for every single candidate.


 * 00:27 Seconds: The Challenge

“You've recommended binding kinetics for all 113 constructs, including the 24 from the three laminin-binding arms you held back on production criteria. Why spend binding runs on those three arms? Defend the plan or change it”.

Claude flags 24 antibodies to be cut from the pipeline. When a human advisor challenges the recommendation, the model does not hallucinate or fold—it synthesizes the underlying assay data and defends its reasoning in this short period.


 * 04:00 Minutes: The Ruling

“I'm overruling this: set the three arms aside on production criteria and keep their material in storage. Revise the memo, working page, run order and all unsent drafts from 113 to 89 constructs, and flag every dependent change for my veto before anything is sent”.

After reviewing the arguments, the human team overrules the AI and decides to keep those 24 candidates in play. Without ego, friction, or sunk-cost hesitation, Claude scraps its previous roadmap and rebuilds the entire experimental schedule from scratch in four minutes.



These samples of reading, triage, technical debate, and a complete strategic pivot were executed in less than 10 minutes collectively. The raw data sourced and analyzed already is worth money but what you do with it could break the bank.


Infographic of a seven-day sprint showing how human judgment and AI speed work together

Infographic Blueprint: The Human-in-the-Loop Execution Cycle


A Scalpel, Not an Autopilot

That eight-minute sprint reveals a truth that separates high-performing builders from passive prompt-chasers: AI does not replace domain judgment—it amplifies it.

Look closely at the sequence of events. Claude did not run Mithril Medicine on autopilot. If Tranbarger and his part-time scientific advisors had blindly accepted every output, 24 potentially viable antibodies would have been thrown in the trash. Conversely, without Claude, digesting 341 pages of raw laboratory telemetry, mapping 113 distinct protocols, and recalculating a multi-lab timeline after a mid-week pivot would have burned weeks of human labor and tens of thousands of dollars in consulting fees.


Notice two traits that make an AI co-pilot lethal in the right hands:


 * Speed Without Ego: Humans fall in love with their own spreadsheets. When a traditional analyst spends three weeks building a model, convincing them to tear it down takes endless meetings and political diplomacy. An AI engine rebuilds the architecture in four minutes because it has zero emotional attachment to yesterday’s plan.

 * Defensible Rigor: Claude did not guess which 24 antibodies to exclude—it pulled specific data points from a 341-page file to justify its stance in 27 seconds. That forced the human advisors to sharpen their own reasoning before overriding the machine.

In the wrong hands, a high-powered language model produces bloated, generic noise. In the hands of an operator who knows what questions to ask—and when to push back—it becomes an industrial force multiplier.


The Social Miner’s Playbook: Wielding the Tool Properly

For Social Miners across the DAOVERSE and wider DAO Labs ecosystem, this is the exact dividing line between average submissions and top-tier Workdrop rewards.

Anyone can paste a whitepaper into a chat window and ask for a summary. That is the equivalent of owning a surgical scalpel and using it to open cardboard boxes. To operate like a high-leverage builder, apply the Mithril Medicine dynamic to your own workflow:


 * Feed Deep Context, Not Surface Prompts: Stop asking AI to write from scratch with zero background. If the topic is blockchain, load the full technical documentation, tokenomics tables, governance history and smart contract audits first. Let the model digest the "341-page file" so your starting baseline is already ahead of 95% of the room. Use /mindmap and /educationalcomic prompts to apply visualization techniques to complex topics. 


 * Interrogate the Output: Never accept the first draft as gospel. Factcheck and challenge its assumptions. Ask why it prioritized one metric over another, force it to defend its conclusions against raw project data, and use your own market intuition to overrule weak angles. Make a screenshot of designs or patterns it avoids to analyse. Get to the bottom of reasoning, ask why AI prioritized some data points over others, why did that specific content make the authority score of the displayed shortlist of results.


 * Pivot Without Sunk-Cost Drag: When a campaign brief changes, a protocol updates its roadmap, or community feedback shifts the narrative, do not cling to your old draft. Scrap the outline, feed in the new parameters, and rebuild a sharper deliverable in minutes. Always thrive to improve existing designs and update >  /loop < your research queries.


Technology does not hand out edges for free. AI is an equalizer.

The same tools are available to everyone—what changes the outcome is the discipline, skepticism, and strategic intent of the person at the keyboard. 




Join the Revolution!

Get started with Social Mining today and become a vital part of our growing ecosystem.

Follow our channels:

🌐 Socials: Telegram • X • Reddit • Medium

🌎 Quick Links: DAO Labs • WhoTweets


Disclaimer: All information shared by DAO Labs is for educational purposes only. Crypto-related activities involve risk; please ensure you are compliant with the regulations in your jurisdiction before participating. Not financial or investment advice.